I've been experimenting a lot with local AI models and figuring out how to make them actually useful. 

I am a hermit and need a new hobby. This works well for me, a former geek and serial obsessive. 

I've discovered a lot, both in terms of what is and is not useful for non-computer-types (I am decades behind in competence when it comes to coding and far, far too lazy and busy to learn modern coding, and the level of sophistication it would require to make something that is useful is far above my level of competence). 

I can do this because I just updated my computer to the latest MacBook Pro, and I loaded it up with memory. I got the super cool one with the OLED screen and 64 GB of memory, giving me headroom to play with mid-sized models while keeping the computer useful for my work. 

First things first: the technology on a modern, vaguely affordable Mac (depending on your budget and how you decide to pay for it) is ready for prime time, but only if you are very clear about what you want it to do, have realistic expectations, and have another AI to help you set it up. Unless you want a basic chatbot that is inferior to what you can get on the cloud, it is pointless unless you enjoy the pure spectacle of watching a local model do vaguely competently and slowly what you can get for free online, or for $20/month if you use it a lot. 

Recommended
Stephen Miller: Superstar Larry Elder Here's Why the Flight 15 Booster Catch Was Abandoned (Update: They Did It!) John Sexton

I couldn't have set it up to do what I wanted without Muse, and I couldn't have defined realistic, useful goals without it either. I would have spent months of frustrating tinkering to get where I am in a weekend with relatively few hiccups. And now I have it doing several tasks that would genuinely cost me a lot of money to run in the cloud, and best of all, it feeds the data to Muse when I need it so the information it collects can be analyzed at a higher level when I need it. 

I built a hybrid local/cloud system that I can keep improving over time, both as I tweak it and as local models get more competent, which they do nearly as fast as they do online. I have no NEED for privacy, since I don't have medical or financial information regulations to worry about and I am not planning to do anything vaguely illegal or alarming, because I am boring as all get out. 

So what do I have it doing locally? I can have it monitor about 50 X feeds at a regular, frequent cadence during working hours, flag trending stories in areas of interest, monitor RSS feeds, sort stories of interest, notify me if something gets flagged, and keep a running tally of stories I might want to write about. Muse built a "Space" that tracks them all by category and expires them after a set time if I don't choose to write about them. 

For X posts with videos, it transcribes and summarizes them, then builds a link database I can search by topic, replacing X's execrable bookmark system. All of these tasks are genuinely useful and are done competently because they are limited in scope and well within the competence of medium-sized models, which are genuinely good but not frontier level.

Still, unless you are a coder, for most people who use AI as a chatbot and not often enough to spend more than $20/month, local AI will be much more expensive than what is available for free or cheap, because the hardware requirements are wildly expensive, and the setup for useful systems like this is difficult. If your goal is anything similar to what cloud AI like ChatGPT, Claude, Grok, or Gemini do for you in quickly searching the web or answering basic questions, my advice is to stay away. Unless you need a complicated and frustrating hobby or want to become a Python expert. 

Don't watch the YouTube videos that hype "Replace ChatGPT at home!" and get excited. It's simply false for 99 percent of people, and don't try to do it without a sophisticated online AI system to help you. Clearly define tasks before you start, or it is a waste of time, and don't expect it to be a cost-free replacement for your AI subscription if you have one. It is a solution for the 0.1 percent right now, and only as an adjunct to the best systems out there. 

That said, I am glad I did it. It genuinely helps me filter the flood of information I subject myself to. But I am a columnist who writes six pieces a day, 5 days a week. 

I don't know how many are tempted to try it, and maybe you need a hobby and aren't looking for something truly useful. When we were young, many of us played with computers, and if you loved diving in headfirst and aren't impatient, it can be fun. 

But for most of us, best to stick with the cloud models. I personally LOVE Muse because it is a great fact-checker, sparring partner for columns when I have time, and has created routines that are genuinely helpful. So far, Zuckerberg is ridiculously generous with usage, although that could change at some point, as it has with other AIs. 

Don't be tempted to let it replace your own thinking, as many people do. Use it as an Executive Assistant that can be trained up to become super-competent over time and personalized to your needs. If you just ask questions occasionally, any assistant can do the job well. 

If you want to build an Executive Assistant to sort your mail, clean out spam, help you find ways to save money, and track your subscriptions and cancel the ones you don't use, I recommend Muse unreservedly. It is smart, adapts to your needs, suggests new things it can do, and is well designed. If you don't want your data used for training purposes, you can opt out too. 







































BEST OF THE BABYLON BEE



BEST OF THE REST:



AND FINALLY...